720p Better Path To 1080p Than 1080i
Dear Gary:
As a Widescreen Review reader since Issue 3 (1993), I congratulate you and the WSR staff for vision, persistence, and a job well done. But Issue 100 (September 2005) has one item stuck in my craw.
While reviewing the Faroudja DILA 1080pHD projection system, Greg Rogers raved about 1080i for its “pixel perfection” to the point of implying 1080i is a better path to 1080p. I don’t doubt that Greg experienced a better 1080i to 1080p conversion with that system, but suggesting 1080i is a better path to 1080p in general, that’s a stretch. I’m sure Joe Kane would echo my point.
I see the Faroudja converting 1080i to 1080p better than it converts 720p to 1080p as an engineering flaw. Why are there 2 to 3 pixels of edge outlining around each vertical and horizontal line for 720p scaled to 1080p? The make-up DCT logic blocks suggest that it should be easier to scale from a progressive image to a progressive image. Progressive scan signal format and displays are also a smarter standard to support for business reasons, given that most HDTVs sold today and in the future will be 720p and 1080p scan displays, not interlace scan displays.
Lastly, I have compared 720p to 1080i sports broadcasts on both 1080i and 720p displays. 720p at 60 fps smokes 1080i at 30 fps. With 720p60, I can actually watch games without the constant distraction of MPEG artifacts. And the smoother more lifelike motion of 720p60 sports broadcasts is even noticeable to my wife and kids.
Thomas Dorsey
mailto:publisher@soulofamerica.com
Technical Video Editor Greg Rogers Comments:
I didn’t suggest or imply that 1080i is a better path to 1080p “in general.” However, 1080i is a much better way to deliver video from a 1080p/24 film transfer than 720p. As I will explain below, it is possible to recover the full 1920 x 1080 pixel resolution of a 1080p film transfer from 1080i, but 720p is forever limited to only 1280 x 720 pixels. The 720p format provides 60 frames-per-second (fps), but the standard film frame rate is only 24 frames-per-second, so that is actually a disadvantage for film. A 720p/24 distribution format would likely be used to prevent judder on film transfers.
A 1080i high-definition film transfer is created by a telecine process using 3-2 pulldown. That means each 1080p frame is separated into two 1080i fields that contain either the odd or the even lines from the 1080p frame. One field (alternately an odd or even field) is repeated after every two frames to convert the 24 frames-per-second film transfer to 60 fields-per-second video. Hence, either 3 fields or 2 fields are alternately extracted from each frame, which is the origin of the term 3-2 pulldown.
A video processor with 1080i inverse-telecine deinterlacing, such as the Faroudja DVP1080, can identify the odd and even fields that originated from the same 1080p frame. It then merges those fields back together to recreate the original 1080p frame without adding any artifacts. It also discards the extra duplicate fields added by the 3-2 pulldown, so the result recreates the 1080p film transfer. However, since this is now 1080p 24 fps once again, the processor (or the projector itself) must convert the video to a progressive frame rate that the projector can display. This is typically 60 fps, but it may also be 48 fps or 72 fps. These conversions are done by simply repeating the frames twice for 48 fps, three times for 72 fps, or alternately 3 and 2 times for 60 fps. Since the latter produces a motion stutter (called judder), it is better to display the video at 48 fps or 72 fps. The 72 fps rate is often used by CRT front projectors, and some fixed-pixel projectors will now display at 48 fps in addition to 60 fps.
High-definition DVDs are expected to store 1080p film transfers in the 1080p/24sf (1080p segmented frame, 24 frames per second) format or the 1080i film format. The 1080p/24sf format is currently used as a professional format to archive high-definition film transfers. The 1080i format for film sources would be similar to the 480i format used by current DVDs. In that case, the 1080i fields are created using the telecine process and 3-2 pulldown, but the “extra” duplicate fields are not actually stored on the DVD. Instead their position is marked in the MPEG stream by flags that tell the DVD player when to repeat a field, and which field to repeat to create a 1080i/60 output. I am hopeful that the 1080i film content will not be vertically pre-filtered for 1080i displays, and that will be a user option on the DVD players. If the film content were pre-filtered, it would reduce the vertical spatial resolution. Although the 1080i film format is stored as interlaced fields, it is stored in a special MPEG mode that is treated as progressive frames for compression purposes. The 1080p/24sf format also divides film frames into interlaced fields, and acts much like 1080i/48. A high-definition DVD player can recreate the original 1080p/24 film transfer from the 1080p/24sf format or the DVD 1080i film format. In the latter case, the repeat field flags from the MPEG stream are simply ignored. Once the DVD player recreates the 1080p24 format, it could output that format as 1080p24, 1080p/24sf, 1080p/48, 1080p60, 1080p/72, or 1080i/60 using the same techniques described earlier for a video processor. I would expect to see many variations of these capabilities to differentiate high-end, high-definition optical disc players in the future. But in initial players, I would expect 1080i/60 and perhaps 1080p/24sf to be output, and the conversion to other formats will be done in video processors or displays. It is also likely that a downconverted 720p output will be available for 720p native displays.
I have said on many occasions that 720p is a better format for live action sports than 1080i. As you have seen, the 60 frames-per-second rate produces smoother motion and, obviously, there are no deinterlacing artifacts. Good deinterlacing of original interlaced video, such as sports and live video broadcasting, requires a much higher level of processing complexity compared to the relatively simple, ideal deinterlacing algorithms used for 1080i (or 480i) film transfers. There are no ideal deinterlacing algorithms for original interlaced video. The 1080i motion-adaptive deinterlacing in the Faroudja DVP1080, and some other processors and projectors, is a good start, but much work remains. Nevertheless, the quality of deinterlacing for original interlaced sources will never equal that of an original progressive source, or a deinterlaced film-source. As I have written before in these pages, I would like to see broadcasters transmit 720p for live video broadcasts, and 1080i film format (or 1080p24) for movies and other programming. That is possible today without waiting another decade for 1080p60.
Your expectations for 720p to 1080p scaling aren’t realistic. No processor can ever scale a 720p source image spatially “pixel perfect” to a 1080p display. In fact the concept of producing a spatially “pixel perfect” image from a source requires that no scaling is done at all. Only a deinterlaced 1080i source, or a 1080p source, could be displayed “pixel perfect” on a 1080p display. As explained in the review, “pixel perfect” means that each pixel from the source is precisely mapped to a single pixel on the display, and each pixel retains its full amplitude. There can never be a one-to-one mapping between the source pixels and display pixels, unless both the source and the display have the same pixel resolution. If 1280 pixels per line from a 720p source are scaled to fit the 1920 pixel line width of a 1080p display, most of the 720p pixels will fall in between the 1080p pixels. The same is true in the vertical dimension. Therefore, scaling is an interpolation function that computes the brightness of the 1080p pixels that surround the location where each of the scaled 720p pixels would lie. Each 720p pixel contributes to the brightness of multiple 1080p pixels, therefore, a vertical line or horizontal line appears to spread out. In addition, the fast rising or falling brightness along edges is like a pulse passing through an interpolation filter. This creates additional overshoot, undershoot, or possibly ringing along the edges. The scaling algorithms are designed to minimize these effects, and the resulting scaling quality is a result of the complexity of the processing and the skill of the designer. As reported in the review, the edge outlining for 720p to 1080p scaling of DVI signals was about two 1080p native pixels around horizontal lines and was negligible around vertical lines. This is exceptionally good scaling performance, and much better than what you mention in your question.
You can E-mail Widescreen Review @ mailto:editorgary@widescreenreview.com